SocialFM: A Social Recommender System with Factorization Machines

摘要

Exponential growth of web2.0 makes social information be an indispensable part for recommender systems to solve cold start and sparsity problems. Most of the existing matrix factorization (MF) based algorithms for social recommender systems factorize rating matrix into two low-rank matrices. In this paper, we propose an improved factorization machines (FMs) with social information, called SocialFM. Our approach can effectively simulate the influence propagation by estimating interactions between categorical variables and specifying the input feature vectors. We combine user trust value with similarity to compute the influence value between users. We also present social regularization and model regularization to impose constraint on the objective function. Our approach is a general method, which can be easily extended to incorporate other context like user mood, timestamp, location, etc. The experiment results show that our approach outperforms other state-of-the-art recommendation methods.